The internal geometry variability of woven fiber reinforced polymer cannot be ignored due to manufacturing process along with many uncertainties, which has nonnegligible effect on mechanical properties such as elastic constants. A statistical analysis of yarn parameters of plain woven carbon fiber reinforced composite was conducted using X-ray micro computed tomography(micro-CT)da- ta. An algorithm based on Correlated Gaussian Random Sequence(CGRS)was developed to construct statistically equivalent yarns of composites, which were further introduced into numerical multiscale modelling approach by establishing Representative Volume Element(RVE)to evaluate the macroscopic elastic properties. The predicted elastic constants showed a good consistency with experimental data by tensile tests, which further showed the importance of considering internal geometry variability to obtain more accurate simulations. In the end, performances of back door of electric vehicle made of the studied composite materials were calculated by finite element model- ling, and excellent prospect of lightweight was demonstrated.